Attribution Evaluation on ImageNet-S
65.5ALRouting Game
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Routing GameBackbone=VGG-16, Config=λ=-1, σ^2=102026.05 | 65.5 | 76.3 | 66.1 | 15.756 | 11.048 | 4.001 | 0.572 | 0.449 | |
| LRP-εBackbone=VGG-16, Config=ε=12026.05 | 63 | 78 | 72.7 | 12.448 | 7.731 | 4.035 | 0.76 | 0.896 | |
| DeepLiftBackbone=VGG-16, Config=γ=0.25, rule=gamma2026.05 | 61.4 | 79.7 | 70.5 | 14.537 | 9.459 | 3.662 | 0.541 | 0.476 | |
| SGBackbone=VGG-16, Config=λ=-2, σ^2=1, Φ̃2026.05 | 61.3 | 73.5 | 68.8 | 16.996 | 13.248 | 4.756 | 1.631 | 0.173 | |
| LRP-γBackbone=VGG-16, Config=ε=0.25, γ=0.12026.05 | 61 | 84 | 68.9 | 12.404 | 7.385 | 3.502 | 0.606 | 0.536 | |
| Routing GameBackbone=VGG-16, Config=–2026.05 | 60.9 | 75.1 | 68.9 | 14.037 | 9.425 | 3.573 | 3.902 | 0.435 | |
| DeepLiftBackbone=VGG-16, Config=–2026.05 | 57.4 | 77.4 | 72.2 | 12.897 | 9.106 | 4.659 | 4.218 | 1.048 | |
| GCAM++Backbone=VGG-16, Config=–2026.05 | 55.5 | 84.6 | 83.8 | 17.301 | 13.728 | 5.697 | 0.504 | 0.58 | |
| IntGradBackbone=VGG-16, Config=Nsteps=502026.05 | 54.5 | 78.4 | 69.6 | 14.452 | 10.511 | 5.293 | 1.275 | 0.975 | |
| LCAMBackbone=VGG-16, Config=–2026.05 | 53.8 | 81.3 | 80.4 | 16.012 | 12.478 | 5.152 | 0.567 | 0.659 | |
| SmGradBackbone=VGG-16, Config=Nsample=30, σ=0.22026.05 | 52.6 | 82.7 | 79.6 | 15.744 | 11.613 | 4.952 | 0.35 | 0.219 | |
| GradientBackbone=VGG-16, Config=–2026.05 | 51.3 | 71.9 | 68.4 | 14.102 | 10.715 | 5.299 | 2.309 | 0.958 |